check hbase

gistfile1.rb

#!/usr/bin/env tumblr_ruby

require 'nrpe_check/extend'
require 'timeout'
require 'json'

@options = {
  timeout:        120, # seconds

  hbck_command:   'hbase hbck 2>/dev/null',
  hbck_keys: [
    'Version',
    'Master',
    'Status',
    'Number of Tables',
    'Number of live region servers',
    'Number of dead region servers',
    'Number of backup masters',
  ],
}

check do
  status :unknown, 'unable to determine hbck status' # register an initial status

  # query hbase for status
  hbck_output, hbck_status = Timeout.timeout @options[:timeout] do
    [%x[#{@options[:hbck_command]}], $?]
  end

  # parse result
  hbase_status = hbck_output.lines.reduce({}) do |status, line|
    if @options[:hbck_keys].any? {|key| line.include? key}
      key, value  = line.chomp.split ':'

      status[key] = value.strip
    end

    status
  end

  # check hbase
  case
  when hbase_status['Number of dead region servers'].to_i != 0
    exit_with_status :critical, "dead region servers: #{hbase_status['Number of dead region servers']}"
  when hbase_status['Status'] != 'OK'
    exit_with_status :critical, "hbck status: #{hbase_status['Status']}"
  when hbck_status.exitstatus != 0
    exit_with_status :critical, "hbck status: unhealthy"
  else
    status :ok, hbase_status.to_json
  end
end